Modelling choice in digital writing: functional revisions and 'texture'

(1)

Modelling choice in digital writing:

Functional revisions and 'texture'.

A thesis submitted in fulfilment of the requirements for the degree of Doctor of Philosophy in Language and Communication

By

Neil Evan Jon Anthony Bowen

First Supervisor: Lise Fontaine Second Supervisor: Gerard O'Grady

School of English, Communication, and Philosophy (ENCAP)

Cardiff University, August, 2016.

(2)

Summary

In this thesis, the digital writing practices of two 2nd_{year undergraduates are examined in terms}

of the functions and structures of their revision activity.

Using systemic functional linguistics as an underlying framework, the project takes a first step toward to a dynamic description of written text in functional terms. To date, research into dynamic descriptions of language (i.e. the logogenesis, or unfolding of meaning in a text) has been almost entirely based upon data related to the spoken mode. Furthermore, research into revision activity has tended to ignore the functionality or meaning inherent in such revisions. The existing research has, instead, primarily focused on cognitive processes (for e.g., pause times) or which language structures, such as parts of speech, are more frequently involved in revisions that others. Ultimately, this thesis works toward providing a dynamic description of the language functions and revisions involved in revision activity in two student writers.

To do this, it makes use of software called keystroke logging to record how two writers compose four academic essays on their computers. Such technology allows us to model the unfolding of a written text in much the same way as a tape recording allows researchers to model the unfolding of a speech. By examining how these writers revise text in light of academic expectations (a 'valued' configuration of field, tenor, and mode register variables present in language choices) and digital mediation (computer afforded composing practices), the thesis shows how certain language functions and structures may play a key role when it comes to shaping an academic essay. In this light, this thesis takes a first step to providing a dynamic description of what is usually analysed solely in synoptic terms, by showing how we can analyse written text as process (an evolving entity) rather than just a product (a static entity). Because of this, a new model of analysis – a combination of keystroke data and functional systemics – is proposed, which can provide an additional perspective to the already existing methods of examining writer behaviour by looking at meaning making practices in revision activity.

(3)

Declarations Page

STATEMENT 1

This thesis is the result of my own independent work/investigation, except where otherwise stated. Other sources are acknowledged by citations and/or footnotes giving explicit

references. A Bibliography is appended.

Signed ………. (candidate) Date: #

STATEMENT 2

This work has not previously been accepted in substance for any degree and is not concurrently submitted in candidature for any degree.

STATEMENT 3

This thesis is being submitted in fulfilment of the requirements for the degree of Doctor of Philosophy in Language and Communication

STATEMENT 4: TO BE COMPLETED WHERE A COPY OF THE DISSERTATION IS SUBMITTED IN AN

APPROVED ELECTRONIC FORMAT

I confirm that the electronic copy is identical to the bound copy of the dissertation Signed ………. (candidate) Date: #

STATEMENT 6

I hereby give consent for my thesis, if accepted, to be available for photocopying and for inter-library loan, and for the title and summary to be made available to outside

organisations.

STATEMENT 5: BAR ON ACCESS APPROVED

I hereby give consent for my thesis, if accepted, to be available for photocopying and for inter-library loans after expiry of a bar on access approved by the Graduate Development Committee.

(4)

Acknowledgements

I owe a great debt of gratitude (and numerous apologies) to my Supervisor, Dr Lise Fontaine. I realise that I am particularly odd (both in how I work and how I am in general), yet despite all of this you have left me to get on with it, which shows how much faith you actually had in my ability to create something half decent. For this I am forever grateful. I also owe a great deal of thanks to the Inputlog team at the University of Antwerp for their continued help and advice. Without which, this thesis would not have been possible, both in terms of its inception and its completion. I would particularly like to thank Professor Luuk Van Waes for the many hours of time he has given up to answer my emails and questions, Marielle Leijten for her enthusiastic support, and the invaluable opportunity to meet people from outside my primary theoretical field.

I would also like to thank the staff at the Centre for Language and Communication Research at Cardiff University who have always been so gracious with their time and linguistic expertise. It seems fitting to thank first the very humble woman who made me realise that averaging 86% in a Bachelor's degree was something to be proud of. So, I extend my deepest thanks to Professor Tess Fitzpatrick, my personal tutor at Swansea University and now the deputy head of CLCR, who has acted as both an academic advisor and therapist on many an occasion -- without your support and kind nature I would not have finished my bachelor's degree, let alone gone on to a Masters and a PhD. I would also like to thank Professor Alison Wray who, from the moment we first met, has shown support and interest that reflects her professionalism and considerable intellect -- your 'extensive' feedback on my many research proposals (and I mean that in the nicest way) is just one example of the many ways you have encouraged and challenged me to improve the way I think and express myself. Finally, I would like to thank Dr Gerard O’Grady (my second supervisor) and Dr Tom Bartlett for their illuminating insights and encouragement to all things SFL related, as well as for entertaining my many off topic ramblings and divergences that quite honestly had nothing to do with my thesis.

(5)

List of figures and tables

Figure 2.1: The Hayes-Flower model of the writing process 9

Figure 2.2: Bereiter and Scardamalia's Knowledge-transforming model 11 Figure 2.3: Hayes' new(er) model of the writing process 13

Figure 2.4: Hayes' latest model of the writing process 14

Figure 2.5: Macro-stages in the cognitive development of writing skill 16 Figure 2.6: Interdependence of language, knowledge, and memory during writing 18

Figure 3.1: Relation between semogenic processes 30

Figure 3.2: SFL’s stratified view of language 32

Figure 3.3: SFL's cline of instantiation 34

Figure 3.4: Grammatical metaphor as stratal tension 38

Table 3.1: Ways of calculating lexical density 41

Figure 3.5: Voice and communicative stance in academic writing 43

Figure 3.6: Axis by which stance is typically analysed 45

Figure 4.1: Basic elements of comprehending language 54

Figure 4.2: SFL's cline of instantiation 54

Table 4.1: Discursive genres (or ‘modes of thought’) 55

Figure 4.3: Bazerman’s (2004) communicative pyramid 56

Figure 4.4: Integration of genre knowledge 58

Table 4.2: Two types of uptake via intermediary genres 60

Table 4.3: A categorization of written texts and their disciplines 64

Table 4.4: Distribution of genre families 65

Figure 4.5: Distribution of genre families across disciplines 65

Table 4.5: Genre family: Essay 66

Figure 5.1: Interactions amongst methodological sections/concerns 73

Figure 5.2: SFL's stratification of language 78

Table 5.1. Constituency (or rank scale of analysis) 79

Figure 5.3: Social context and meaning realized through metafunctional strata 80

Table 5.2: Metafunctional agnation 81

Figure 5.4: The system of POLARITY 82

Figure 5.5: Information flow in the English clause 90

Figure 5.6: Coding system for textual analysis 90

Table 5.3: Table of lexicogrammatical systems 93

Figure 5.7: Function-rank matrix 94

Table 6.1. General overview of document construction (JD1) 98

(10)

ii

Table 6.2: Frequency of revisions types (JD1) 100

Table 6.3: General overview of document construction (JD2) 103

Table 6.5: General overview of document construction (JD3) 104

Table 6.7. General overview of document construction (BB) 105

Figure 6.2 Focus events (BB) 106

Table 6.8: Frequency of revisions types (BB) 107

Table 6.9: Distribution of FPs in each dataset 109

Figure 6.3: The unfolding of revision types in session 1 for each dataset 109

Figure 6.4: The unfolding of revision types in JD1 112

Figure 6.5: The unfolding of revision types in JD1 112

Figure 6.6: Percentage of revisions in each session that were FPs (JD) 113 Figure 6.7: Percentage of revisions in each session that were FPs (JD) 113

Figure 6.8: The unfolding of revision types in BB 114

Figure 6.9: Frequency of revisions types in the middle session of each dataset 115

Table 6.10: Percentage of revision types that were INSAs 115

Figure 6.10: Percentage of revisions that are INSAs in each dataset 116 Figure 6.11: The partitioning of a T-unit in terms of THEME/RHEME components 117 Figure 6.12: The distribution of revisions in terms of their realization 118

in THEME/RHEME components

Figure 6.13:Participant tracking in English 120

Table 6.11: Subject Themes and their revision activity in paragraph 1, JD1 121 Figure 6.14: Percentage of revisions in each session that were INSBs 123 Figure 6.15: Percentage of revisions in each session that were CPs 125 Figure 6.16: Distribution of revision types across JD's dataset 128 Table 6.12: No. of words typed (process counts) and no. of words in each 129

text (product counts)

Figure 7.1: Using a function-rank matrix to narrow down systemic choice 134

Figure 7.2: Functional choice in JD1's revisions 135

Figure 7.3: Functional choice in JD2's revision 135

Figure 7.4: Functional choice in JD3's revision 135

Figure 7.5: Functional choice in BB's revision 135

Table 7.1: Frequency of revisions in functional terms 136

Figure 7.6: Rank realizations in JD1's revisions 139

Figure 7.7: Rank realizations in JD2's revision 139

Figure 7.8: Rank realizations in JD3's revision 139

(11)

iii Table 7.2: Frequency of rank level realizations in revision based activity 139 Table 7.3: Function/rank matrix for revision activity in JD1 141

Table 7.4: Revision activity and systemic choice in JD1 142

Figure 7.10: Distribution of revisions in terms of systemic choice (JD1) 142 Table 7.5: Top ten systems/functions/ranks affected by JD1's revisions 143 Table 7.6: Function/rank matrix for revision activity in JD2 143

Table 7.7: Revision activity and systemic choice in JD2 144

Figure 7.11: Distribution of revisions in terms of systemic choice (JD2) 144 Table 7.8: Top ten systems/functions/ranks affected by revision activity in JD2 145 Table 7.9: Function-rank matrix for revision activity in JD3 145 Table 7.10: Revision activity and systemic choice in JD3 146 Figure 7.12: Distribution of revisions in terms of systemic choice (JD3) 147 Table 7.11: Top ten systems/functions/ranks affected by revision activity in JD3 147 Table 7.12: Function/rank matrix for revision activity in BB 146

Table 7.13: Revision activity and systemic choice in BB 148

Figure 7.13: Distribution of revisions in terms of systemic choice (BB) 148 Table 7.14: Top ten systems/functions/ranks affected by revision activity in BB 149 Table 7.15: Top five systems involved in revision activity 149 Figure 7.14: Orbital (mono-nuclear) structure of experiential meaning in the NGrp 151

Table 7.16: Nominal systems in English 152

Figure 7.15: System of DETERMINATION 157

Figure 7.16: The systems of clause complexing 160

Table 7.17: Summary of choices in interdependency relations for complexes 161 Figure 7.17: Unfolding interdependency choices in revision activity (JD1) 162 Figure 7.18: Unfolding interdependency choices in revision activity (JD2) 162 Table 7.18: Revisions contributing to NGrp complexity in each dataset 166 Table 7.19: Number of revisions involving GM in each dataset 169 Figure 8.1: The unfolding of language functions in revision activity (JD1) 173 Figure 8.2: The unfolding of language functions in revision activity (JD2) 176

Table 8.1: Breakdown of interpersonal revisions in JD2 177

Figure 8.3: The unfolding of language functions in revision activity (JD3) 178

Table 8.2: Breakdown of interpersonal revisions in JD3 179

Figure 8.4: The unfolding of language functions in revision activity (BB) 181 Figure 8.5: Comparison of functional uptake ratios across the four datasets 182 Figure 8.6: The unfolding of rank realizations in revision activity (JD1) 184 Figure 8.7: The unfolding of rank realizations in revision activity (JD2) 188 Figure 8.8: The unfolding of rank realizations in revision activity (JD3) 189

(12)

iv Figure 8.9: The unfolding of rank realization in revision activity (BB) 191 Figure 8.10: Uptake ratios of rank constituents via revision activity in each text 192 Figure 8.11: Unfolding expansion and projection in revision activity (JD1) 194 Figure 8.12: Unfolding expansion and projection in revision activity (JD2) 200 Table 8.3: Uptake of expansion and projection in JD2's revisions 201 Figure 8.13: Unfolding expansion and projection in revision activity (JD3) 203 Figure 8.14: Unfolding expansion and projection in revision activity (BB) 202 Figure 8.15: Uptake of expansion and projection in revision activity across all four texts 206

(13)

v

Conventions

Systemic description

Case

Convention

Example

lower case

name of system term (option or feature)

intensive, identifying process

small capitals

name of lexicogrammatical system

THING TYPE, THEME, MOOD, QUALIFICATION

initial capital

letter

name of functional element realized by a

systemic choice

Thing, Theme, Mood, Qualifier

Operators used in systemic realization statements/descriptions

Description

Operator

Example

Choices selected within a system

[systemic choice]

Choice in

_process]

EVENT TYPE

= [Relational

Structural realization of functional

element

(structural unit)

[Relational process (verb)]

Movement through a system from

left to right (increasing delicacy of

description)

:

(1st_{layer :}₂nd_{layer : 3}rd_layer)

[comment Adjunct (adverb):

propositional: on whole: assertive]

systemic combination of two system

(14)

vi

List of Abbreviations

ActLits - Academic Literacies AE - Academic English

BAWE - British Academic Written English corpus CP - Commutative Progression

ESP - English for Specific Purposes FP - Forward Progression

GM - Grammatical metaphor INS - Unclassified revision INSA - Forward Insertion/Deletion INSB - Backward Insertion/Deletion KSL - Keystroke Logging

LTM - Long-term Memory MDA - Multidimensional Analysis NGrp - Nominal Group

NR - New Rhetoric

WM - Working Memory

RQ - Research Question

SFL - Systemic Functional Linguistics STWM - Short-term Working Memory VGrp - Verbal Group

(15)

1

Chapter 1 Introduction

1.1 Overview

In the past two decades, computers, and the innovations they have spawned (particularly the internet), have transformed how we interact (e.g. social networking), do commerce (e.g. online auctions), stay informed (e.g. RSS feeds), and, most significantly for this thesis, how we search for, access, and present information. These advancements have (and are) changing the way we produce, disseminate and consume texts. One particular sphere where this has become evident is the world of student writing, where students now have the means to travel between physical and virtual worlds, pushing and pulling information through spatial and acoustic environments they both partake in and (co)construct. Oblinger and Oblinger (2005) posit that this ‘net’ generation build understanding by drawing on multiple sources, leading to non-linear thought processes based on visualizations. Although, Oblinger and Oblinger’s treatise is premised on Western adolescents—Eastern students may prefer different modes of learning depending on their socio-cultural heritage—their view is applicable to this thesis because here we will only be examining Western (L1 English) students. However, the traditional view of 'literacy' privileges alphabetic forms, words, and their sequences—a document-centric view that promotes a start-to-finish view of knowledge transmission. This is somewhat at odds with the increasing prominence of digital text, where what constitutes a 'page' is radically transformed via hyperlinks, hypertext, and 'live' information (Kress, 2003, p.3). Moreover, as Trupe (2002) convincingly argues, digitally composed text(s) draw on new ways of writing, where single mode genres are fast becoming obsolete; hence, the recent interest in multimodal or hybrid-based views on text (cf. Domingo, Jewitt, & Kress, 2015). Ultimately, modern-day writers seem to be immersed in the 'mediated multimodal genre system' (Prior, 2009).

This shift toward a multimodal, dynamic workspace has spurred renewed interest in the concept of 'literacy', and has unknown implications for our understanding of how students' compose texts. For example, initial investigations point to growing instability within genre systems, where modes of meaning-making are made and remade in a matter of months (Bowen & Whithaus, 2013; Domingo et al., 2015). For example, in this multimodal, dynamic workspace, students can represent themselves (and their knowledge) in a myriad of ways, ranging from the more traditional words on a page approach (common in most essays), to embedded charts, images, sounds, and hyperlinks (typical in PowerPoint presentations, for example), and the ability to use motion, zoom, and spatial arrangements to set up links between ideas/content (e.g., Prezi). In such dynamic workspaces, ideas and the relationships between them may be more representative of the producer's actual thought patterns, which will probably, more often than

(16)

2 not, be non-sequentially organized (which can be limiting factor in writing). For example, whilst the ability to copy, paste, and cross-reference information endows digital text with increasing fluidity and temporariness, the new modes of meaning-making (some of which were listed above) draw much more heavily on visualizations and projected (or guided) physical movements, which seemingly mirror more closely the elasticity with which we hold ideas in memory (Olive & Passerault, 2012). In writing, though, these networks of non-linear thought patterns need to be transformed into linear, sequential arrangements, where the affordances listed above get reduced to the organization of text and 'pointers' within that text that tell us (the reader) where to look for such connections and related ideas/content. In writing, then, the writer is effectively reorganizing a collection of 'floating' ideas/content into an unfolding collection of words. It is in this transformative process that we find a sequence of interactions between the writer and their text, or as Prior (2009) puts it:

'the composed utterance has a history where a sequence of interactions and possibly a series of externalized inscriptions have been organized around the project of a final text/performance.' (p.27)

This history is particularly rich if we take into account computer mediated composition, where the writer can interact with and manipulate unfolding text using digital applications (drawing packages, spreadsheets, etc.), digital sources (worldwide web), and more traditional note-taking techniques (digital and/or pen-and-paper). It is perhaps unsurprising, then, that a great deal of a writer's time can be spent searching for, and switching between, multiple sources, making use of ‘old’ text in the pursuit of ‘new’ (e.g., Leijten, Van Waes, Schriver, & Hayes, 2014; Swarts, 2010). However, in the majority of writing research, text is perceived as equivalent to speech, as if it was conceived of and realized in a single instance, rather than the product of hours, days, weeks, perhaps even years of agonizing deliberation, reflection, revision, re-organization, and finally production. As Prior (2009) highlights:

'Even in some of the richest theoretical and empirical work, there remains a tendency to freeze writing [...] to see writing as a noun rather than a verb.' (p.22)

For example, although O'Donnell (2013), and others (e.g., Coffin & Donohue, 2012), frequently highlight that Systemic Functional Linguistics (henceforth SFL) has the potential to examine written text as process (dynamic descriptions of paradigmatic choice), SFL research into text as process has concentrated almost exclusively on speech1_{(e.g., O'Donnell, 1999; Ventola, 1987;}

Yang, 2010). Furthermore, most linguistic-based research into writing has been decidedly retrospective, focusing on synoptic descriptions of syntagmatic realizations (no doubt due to the

1_{O'Donnell (2013) rightly notes that investigations into speech are fundamentally different from writing}

because you cannot change a previous utterance. However, you can rephrase something you have just said, much like you can revise what you have just written.

(17)

3 difficulty of analysing paradigmatic choice as it unfolds (cf. Ventola, 1987)). This is especially evident in Australian ‘Genre Theory’ (cf. Chapter 4), where an academic text is seen as ‘a staged, goal-oriented social process realised through register’ (Martin, 1992, p.505). Similarly, in writing process research, the notion of written text is also centred on the conception of a finished product, only the primary focus here is on the cognitive processes involved in writing (cf. Chapter 2). Therefore, this study takes up the challenge of studying the language of written text as process by examining what academic writing in the digital sphere has come to represent. More specifically, the thesis focuses on how two student writers shape the linguistic features (or meaning-making potential) of their texts in real-time. It does this by assuming that unfolding language choices in written text can be examined in part by exploring how students revise text in terms of choices at the morpheme level and above (i.e. the rank units related to lexicogrammatical choice, which will be outline in Chapter 5, §5.2.2.3). I have limited my examination to two participants because: (a) the current state of automatic analysis of linguistic features during the writing process is limited, resulting in the need for extensive manual coding; (b) there is no established methodology with which to examine the function of writer's revisions, and (c) because of (a) and (b) a limited sample means that I can provide a thick description of an hitherto unexamined area, allowing for both methodology and theory to be explored more fully.

1.2 Research questions and the underlying motifs used to address them

To explore this previously unrealized thread of research into written text as process, this thesis subsumes a number of underlying motifs (or themes), which are reflected in three sets of research questions (RQs):

1. Development of how students write:

a. What practices do students use when digitally composing text?

b. When (sequentially) and where (within the clause) are these practices employed?

c. Which practices (if any) are relatively stable, and which appear to change over time?

d. Do 'good' writers converge on similar practices? 2. Development of

what students write:

a. What are the key linguistic features of 2nd year undergraduate revisions? b. Are these features comparable between/within different writers? c. Are these features comparable to those of more experienced writers? 3. Product/process

relationship(s)

Is there a relationship between how a person writes and the perceived quality of their text(s)? I.e. does the process affect the product?

These three sets of RQs are broad in scope and cover more than one theoretical concern or thread of research. Because of this, I have organized the thesis into three separate, yet complementary areas of investigation (the study's three underlying motifs), each of which reflects one of the three RQs listed above, and each of which will have its own literature review chapter and discussion

(18)

4 chapter (cf. §1.3). Naturally, there will be some overlap in certain chapters, but where this occurs, I have attempted to outline these commonalties and interconnections as clearly as possible. Before introducing these three underlying motifs, I should make it clear that this study takes the view of ‘discourse as genre’ (Bhatia, 2004, p.20), where the primary concerns are the linguistic, socio-cognitive, and ethnographic view of texts, which can be broadly related to three conceptual spaces: (i) text, (ii) context, and (iii) society/culture, where analysis moves from context to text in 7 steps (Bhatia, 2004, pp.153-182):

1. The text is placed in its given situational context (identify genre). 2. Existing research on the genre is surveyed (literature review). 3. Goals and relations of genre to readers/writers is explored. 4. Primary corpus is collected (product and process data).

5. The context surrounding production/reception is investigated (research site). 6. Texts are analysed for linguistic features (data analysis and discussion).

7. Informal feedback from participants is used to inform analysis (member reflections). Following these steps provides a ‘thick’ description of the text, context, and persons involved. Therefore, using this overarching assumption as the thesis's base, I present the following three underlying motifs (explained in the following subsections) as a first step toward addressing the three sets of RQs outlined above. Following this, these three motifs (and their accompanying RQs) will be picked up in §1.3, where I outline their relevance to the organization of the thesis in terms of their contribution to each chapter.

1.2.1 How students write

The first theme of the thesis sets out to address the question of 'how' students write. To do this I make use of keystroke logging software to capture the writing activity of two students as they composed 4 essays as part of their undergraduate assessment (more will be said of this technology later): one student produced three essays over two months, the other student produced one essay over two weeks. Using data collected via keystroke logging software, I explore what these student writers did when digitally composing text, such as their normal text production (non-interrupted bursts of activity), their 'focus events' (sources and the programs they consulted/used outside of MS Word), and their revision activity in terms of the when and where of revision 'types' (set out in Chapter 5, §5.6).

1.2.2 What students write

The second theme of the thesis sets out to address the question of 'what' student write. To do this, I use an experimental mixed methods approach, which combines keystroke logging data with

(19)

5 a traditional text linguistic analysis (SFL based). This approach allowed me to build up a dynamic description of text as it unfolded. This is a departure from the normal practice of analysing a finished text (a synoptic description), as we shall attempt to reveal (and examine) the language choices a writer makes as he/she writes, and not just those that are visible in the final draft. This means examining deletions and insertions as they occur in the process of text construction, particularly in terms of how they contribute to the language functions/structures of each text. Specifically, we will look at the realization of unfolding choice in terms of SFL's views on language function, rank, and systemic choice (explained more fully in Chapter 5).

1.2.3 Product-process relationships

Finally, in an effort to be 'appliable' (Halliday, 2009), the third theme of the thesis aims to examine the product in relation to the process. More specifically, it will attempt to examine if the process of text construction can be linked to its finished 'quality' in any way, and, if so, are there any theoretical benefits to be gained from such an endeavour? Studies in this vein are very rare, no doubt due to the complexity involved in drawing such connections (as we shall see in the next chapter). Therefore, this final theme is more exploratory and takes a somewhat subsidiary role to the other two themes; as a consequence of this, the final RQ is much broader in scope and less specific in its aims. Nonetheless, it is an important endeavour, and is one that will run implicitly throughout the thesis, as well as being explicitly covered in the Chapter 8.

1.3 Organisation and contribution of thesis

As already noted, the three central motifs outlined above broadly inform the three sets of RQs that provide the underlying aims of this thesis. Therefore, in organizing the thesis I have chosen to give each motif its own literature review chapter. Accordingly, Chapters 2, 3 and 4 cover the literature related to each of the motifs in turn. Although, these three chapters appear as separate entities, they are in fact connected in that they cover: (a) motif number 1: what writers 'do' (Chapter 2); (b) motif number 2: what academic writers produce (Chapter 3); and (c) motif number 3: what academic writing 'does' (Chapter 4). More specifically, these 3 chapters explore the main theories and research behind what writers generally do (in terms of underlying cognitive process), what academic writers are expected to produce (typical linguistic features of 'model' academic texts), and how academic texts come to be situated within larger socio-historical contexts as genres and genre families (how academic texts relate to each other and their audiences). As such, these three chapters are necessarily broad in scope, and address literature in several disciplines: cognitive psychology, linguistics, rhetoric, and social-realist education. This interdisciplinary focus is warranted due to the integrated nature of writing and writing

(20)

6 development. Because the literature in these fields is extensive and broad in scope, in order to draw connections between the three literature review chapters, I have chosen to adopt the view that writing involves a complex interplay of five types of knowledge (Hyland, 2011, p.31):

1. Content knowledge: (Re)appropriating/positioning knowledge in terms of dialogism (Hyland, 2005, 2008; Tang, 2009) and heteroglossia (Biber, 2006; Hood, 2010; Scott & Turner, 2009);

2. Contextual knowledge: Awareness of activity systems (Bazerman, 2013), and wider societal/cultural expectations (Martin & Rose, 2008);

3. Genre knowledge: Knowledge of generic conventions and depth and breadth of genre networks (Nesi & Gardner, 2012; Tardy, 2009);

4. System knowledge: Learning how to use uncommonsense linguistic repertoires (Christie, 2012; Martin, 2007);

5. Process knowledge: Movement from intra- to inter-level concerns (Kellogg, 2008), in combination with learning-through-writing (Galbraith, 2009).

By using these five types of knowledge, I hope to situate the contents of each review chapter (2, 3, and 4) in relation to each other, and to the field in general, creating a kind of overarching theme that informs the rest of the thesis, its findings, discussions and, ultimately, its conclusions. In relation to the five knowledge types outlined above, then, item 1 (content knowledge) is somewhat taken for granted in this thesis, because both writers have average grades that subsume their level of content knowledge to be high (both average over 70% in their studies). To cover items 2-5, however, I make a distinction between the writer (process knowledge covered in Chapter 2), the text (system knowledge covered in Chapter 3), and the socio-historical context behind writing (genre theory covered in Chapter 4). The ordering of these chapters reflects somewhat the order of the RQs outlined above, and the discussion chapters that follow. I.e. the ordering in no way reflects the importance of one strand of research over another, it is simply a means to simplify the organization of the thesis as much as possible in light of a complex 'research problem'.

Chapter 5 outlines the methodology used to address this 'research problem'. Its organization attempts to align with Tracy's (2013, p.230) eight 'big tent' criteria of excellence in qualitative research by addressing the following concerns: research design, sampling selection, research site, data collection, validity and reliability, data analysis, researcher bias and assumptions, and ethical considerations. Although these concerns are presented as discrete components (or sections), the chapter repeatedly highlights how they are parts of an integrated whole. Furthermore, the methodology represents a contribution to both writing research and linguistics as it combines

(21)

7 two separate yet complementary methods/frameworks: keystroke logging and text linguistic analysis (based firmly in SFL).

Chapters 6, 7, and 8 present both the results and the discussion of each set of results in relation to each of the three RQs. I.e. rather than include a chapter entitled results, I have chosen to incorporate the relevant results alongside their accompanying discussion(s) so as to address each of the main RQs (or motifs) in turn within a self-contained chapter. Accordingly, Chapter 6 concerns itself with how the two student participants write, and, thus, primarily relates, and contributes, to the research covered in Chapter 2. It seeks to answer RQ 1. How students write? Chapter 7 concerns itself with the development of linguistic features found in text and, thus, primarily reflects upon, and contributes to, the work covered in Chapters 3 and 4; it seeks to answer RQ2. What do students write? Chapter 8 is exploratory, in that it covers the unfolding of language in written text—it deals with what SFL refers to as the 'logogenetic time frame' (covered in Chapter 3). This area is severely under researched and, thus, the work contained in this chapter contributes to the work covered in Chapters 2, 3, and 4 by exploring a perspective of writing that has not yet been undertaken—the unfolding of language choices in written text as it is being produced. It seeks to explore RQ3. Product/process relationships.

Chapter 9 is the conclusion chapter. It brings together the previous chapters and outlines in more detail the overall contributions of the thesis to the three main strands of research covered by the RQs and the literature review. In broad terms, these were research into revision activity, SFL studies into written text, and research into student writing in general. It also outlines the major findings of the thesis, its limitations, and suggests areas for future research, whilst also reflecting on the state of current research.

(22)

8

Chapter 2 Process knowledge: What writers do

Introduction

This chapter focuses on cognitively oriented research into writing, which conceptualizes writing as involving various interrelated processes of differing complexity. Thus whilst Chapter 3 will examine connections between features of writing (products), and Chapter 4 will examine connections between writing activities (genre theory), this chapter will examine connections between cognitive processes, focusing on three key recursive processes that are central to process-based research: Planning, translating, and reviewing/revising. However, because process research is extensive (Pritchard & Honeycutt, 2006; Smagorinsky, 2006), this chapter restricts itself to the following aims: Section 2.1 gives an overview how writing processes are conceptualised in this paradigm; section 2.2 considers what this paradigm contributes to discussions of high-level writing tasks (e.g., academic essays); section 2.3 examines what research into process-product relations has discovered; section 2.4 considers the current state of Keystroke logging (KSL) research with regard to the linguistics of writing. In relation to these four points (or sections), I will attempt to show that (i) while writing process research has made a significant contribution to understanding the main activities related to writing, there is still much to be done with regard to social and affective factors; (ii) in their current formulation, process models cannot fully account for argumentative texts because they underspecify the translation process; (iii) in investigating product-process relationships, process research is slowly moving toward a more dynamic and multidimensional view of writing; (iv) KSL research tends to focus on the language of writing in terms of parts of speech rather than meaning.

2.1 Process models

To summarise and evaluate every process model would be a monumental task (cf. Alamargot & Chanquoy, 2001, for a book length treatment). Therefore, this section focuses on key contributions, starting with Hayes and Flower's (1980) first, and arguably most influential, box-and-arrow type model. At the outset I acknowledge that many process models (or frameworks) are purposely underspecified, and are meant to be modified and augmented in line with emergent findings (Hayes, 2012). Therefore, any criticisms should be seen as areas for future research, rather than outright flaws.

Hayes and Flower’s 1980 model was based on an information-processing approach, and grew out of the findings of one writer's verbal protocols. It provided a framework (or language of description) for investigating the cognitive processes involved in writing, and in so doing, it moved

(23)

9 research away from a view of writing as a product, and re-envisioned it as a problem solving, non-linear, exploratory, and goal driven endeavour. Hayes and Flower postulated that writing was made up of a number of cognitively driven, recursive activities, centring around: planning, formulating (translating), and revising. Being recursive, these processes interact with and/or interrupt each other at any time; i.e., whilst the labels suggest temporal sequencing, processes occur and reoccur in complex patterns. This model is reconceptualised in figure 2.1, incorporating Hayes (1996) and more modern terminology:

Figure 2.1: The Hayes-Flower model of the writing process (1980) This model contained 3 core modules: task environment, cognitive writing processes, and writer's long-term memory (LTM). The task environment incorporates factors that affect the writing task, including goals and text produced so far. The writer’s LTM contains knowledge about topic, audience, and genre. The cognitive module incorporates planning, text generation (originally termed 'translating'), and revising, which are all embedded within a 'Monitor'. In most writing models (this one included), planning incorporates 3 sub-processes: (1) generating ideas from sources; (2) selecting and evaluating these ideas in line with a topic; (3), organizing these ideas in relation to the text's goal, audience, and generic expectations (cf. Chapter 4). Text generation, meanwhile, facilitates the transformation of ideas into language, and revision involves the evaluation of the resulting text and any resulting transformations in terms of spelling, stylistics, etc. Revision, then, relies on a writer's interaction with text, with studies showing that students with higher working memory spans are generally better revisers than students with lower working memory spans (Piolat, Roussey, Olive, & Amada, 2004). The 'Monitor' represents (albeit implicitly) what is now commonly called 'the central executive'. Consequently, this initial model suffered the same 'homunculus' problem that befalls Baddeley and Hitch's model of working memory1_(Olive,

2012), and while Hayes (2012) makes the following claim, the monitor still seems to represent a

1_{If one process controls the functioning of other processes, what controls that one process?}

MONITOR TASK ENVIRONMENT Planning ➙ Idea generation ➙ Organizing ➙ Goal setting Text Generation

COGNITIVE WRITING PROCESSES Revision Reading  Editing WRITING ASSIGNMENT Topic Audience Motivating Cues TEXTPRODUCEDSO FAR WRITER'S LONG -TERM MEMORY Knowledge of: Topic Audience Writing Plans

(24)

10 magical 'black-box'2_:

'The monitor represented the writer’s predisposition to sequence the writing processes in a particular way. It was not intended to control how those processes were carried out.' (p.373)

During the 80s, researchers refined and added to this conception of writing by focusing on reviewing/revising3_{. This research highlighted how novice writers typically focused on local}

revisions, whereas experienced writers made more global level revisions (Flower, Hayes, Carey, Schriver, & Stratman, 1986; Sommers, 1980). Local revisions are minor changes, either at, or below, the sentence level, and thus typically involve lexical choices, corrections to grammar, spelling, etc. Global revisions are major changes, made at the clausal, paragraph, or discourse level, and thus typically involve the manipulation of information. Much of this research, however, was either premised on writing as a top-down process, or imposed an artificial separation between drafting and revising for experimental purposes. Proponents of dual-processing models, such as Galbraith and Torrance (2004, pp.64-65), for example, would link revision strategies to the writer's preferred writing style, and would class revisions made by writers who prefer top-down strategies as 'reactive', in that revisions seek to bring the text in line with initial expectations (i.e. meeting pre-established goals), and thus would be unrelated to developing understanding and would often negatively affect text quality. Writers who prefer bottom-up approaches, on the other hand, would primarily make 'proactive' revisions, leading to increased understanding, but may not affect text quality (Baaijen, Galbraith, & de Glopper, 2014). This complex interaction between revision and writing styles is highlighted by Wallace et al. (1996) and Midgette, Haria, and MacArthur (2008), who show how very brief instruction (8 and 2 minutes, respectively) can lead to more substantive, macro-level revisions. However, while such findings suggest that less experienced writers lack evaluative criteria for what they write (Hayes, 2004), and who they write for (MacArthur, 2013), it may be that such instruction is limited in terms of transfer/future uptake as it may not be generalizable to other situations (cf.Adams, Simmons, Willis, & Pawling, 2010). In another key model, Bereiter and Scardamalia (1987) argue that at least two process models are needed to account for the differences between novice (knowledge-telling) and skilled (knowledge-transforming) writers: knowledge-telling writers generate text in response to an association (usually prior text or an assignment prompt) in a list-like fashion, where elaboration mainly involves formal features (spelling, etc.). In this ‘stage’, writers transform ideas into text under constraints of topic and genre—it is using writing to 'tell'. knowledge-transforming writers,

2_{Discussions of a monitor/central executive are rare, even in present studies (cf. Olive, 2012).}

3_{Perhaps because of this focus on text modification activities, the processes responsible for translating}

ideas into language remains underspecified, both with respect to how letters, words, phrases, etc., are generated and transposed in real-time (however, see Berninger & Swanson, 1994).

(25)

11 on the other hand, generate text in response to a problem (usually self-generated), where elaboration involves not only formal features, but also meaning—this elaboration is typically achieved by iteratively attending to the text (Lindgren, Leijten, & Van Waes, 2011). The knowledge-transforming stage, then, as per the Hayes-Flower model, frames writing as a top-down, problem solving process, but it conceptualizes expert writing as the ability to formulate and solve problems in terms of two spaces: content (topical) and rhetorical (organizational) space, as per figure 2.2:

Figure 2.2: Bereiter and Scardamalia's Knowledge-transforming model (1987, p.12) In this model, knowledge-transforming is a dialogue between content and rhetoric, where writers shape text in light of an overarching problem, emerging text, discourse, and topic: experienced writers negotiate goals, contexts, and actions in relation to themselves and what they know—it is using writing to 'recontextualise knowledge'. Numerous studies support these fundamental tenets (C. Beauvais, Olive, & Passerault, 2011; L. Beauvais, Favart, Passerault, & Beauvais, 2014), where the consensus is that writers resort to knowledge-telling alone because of one (or more) of five key reasons (Alamargot & Fayol, 2009; Hayes, 2012; Torrance & Galbraith, 2006): (1) underdeveloped language, (2) inefficient writing strategies, (3) lack of topic knowledge, (4) underspecified goals, and/or (5) lack of motivation. However, in relation to argumentative writing, Coirier, Andriessen, and Chanquoy (1999) suggest that:

'two perspectives are playing a role: problem solving/topic structure, and argumentation structure. Experts can solve the dissonance between these perspectives [...]. Non-experts rely mostly on topic structure.' (p.14)

It could also be argued, then, that in constructing argumentative texts, a knowledge-transforming stage would need an additional 'pragmatic/interpersonal space', which could interact with the

Mental representation of assignment

Problem analysis and goal setting Conte nt pr oble m s pace Rhet or ica l pr oble m s pace Discourse knowledge Problem translation Content knowledge Knowledge-telling process

(26)

12 other two spaces in light of restructuring a text (and its contents) toward different goals, points of view, or perspectives. Highlighting that in such texts the points made are sometimes less important than how they are made (Hyland, 2012; Smagorinsky, Daigle, O'Donnell-Allen, & Bynum, 2010). Consequently, one could say that the knowledge-transforming model appears to privilege cohesion over coherence. Furthermore, although Bereiter and Scardamalia (1987) illustrate that knowledge-transforming is only evident in more experienced writers (postgraduates), they fail to explain how one moves from telling to knowledge-transforming or, therefore, what occurs between these phases (Hayes, 2011)4_{. Thirdly, as with}

the Hayes-Flower (1980) model, the distinction between novice and expert belies the fact that writing is field specific, where novices in one field may be experts in another, and vice-versa; i.e. these models explicated a large portion of context, focusing on cognition, to the behest of situational and interactional concerns.

During the 90s, process research shifted toward memory and writing. It further incorporated Baddeley's (2003) model of working memory and in The science of writing: Theories, methods, individual differences, and applications, three new(er) models were proposed:

Firstly, Kellogg (1996; elaborated in 2001) drew on Bereiter and Scardamalia's (1987) model and previous research to produce a speculative model of how writers mature to become experts, moving from knowledge-transforming to 'knowledge-crafting'. Kellogg based his thesis on the logical assumption that experienced writers are able to hold three distinct representations of text in WM: knowledge conveyed by the text, the author's representation of that knowledge, and the author's judgement on how that text will be received (virtual reader's perspective). Experienced writers are believed to be able to juggle cognitive processes with higher-level concerns of content, rhetoric, and audience awareness because they have developed automatized routines that prevent overloading the central executive. Regardless of which view of WM one assumes5_,

two basic mechanisms appear to reduce cognitive demands: (1) automatizing task information (e.g., increasing topic knowledge) or task demands (e.g., using generic templates); (2) using compensating strategies to consciously reduce the load on non-essential activities (e.g. ignoring spelling mistakes). Mechanisms for reducing constraints, then, can top-down (outlining, conceptual maps) or bottom-up (freewriting6_{, extensive revising). Kellogg (2008) also suggests}

that demand can be reduced by: (1) maturation of the central executive with age; (2) committing

4_{Hayes is currently working on an intermediary stage: 'knowledge-structuring'.}

5_{Studies into WM constraints typically assume: (1) Dual-task interference (a capacity theory of writing),}

which posits that when an individual is engaged in two or more tasks, the performance on one, or both tasks will reduce (McCutchen, 2000; Olive, 2012), or (2) a componential view of writing, which believes there are limits to how much information we can store in any one component of memory (Kellogg, 1996).

6_{Freewriting delays revising and editing until after the first draft (a dual-draft approach); see Elbow (2012)}

(27)

13 disciplinary knowledge to LTM; (3) automatizing some aspect of planning, revising, or generating. Similarly, Torrance and Galbraith (2006) suggest: (1) automatizing low-level skills; (2) developing compensatory strategies to reduce parallel processing; (3) minimizing on-line planning by making notes/outlines.

In the same book (The science of writing: Theories, methods, individual differences, and applications), Hayes (1996) focused on the sub-processes related to revision, cognition and affect, and reformulated his original model, as per figure 2.3.

Figure 2.3: Hayes' new(er) model of the writing process (1996, p.4) The major changes in this model are the increased attention to WM, addition of visuospatial sketchpad, integration of motivation/affect, and the reconfiguration of cognitive processes so that text interpretation becomes more central. These changes aimed to account for internal predispositions to writing, language ability, and social/physical influences. However, while the model sees the task environment as essential, it is left underspecified, as is text production and motivation/affect. Some of these issues are carried forward to Hayes (2012):

THE INDIVIDUAL THE TASK ENVIRONMENT

COGNITIVE PROCESSES Text production Reflection Text interpretation LONG-TERM MEMORY Task schemas Topic knowledge Audience knowledge Linguistic knowledge Genre knowledge The Social Environment

The Audience Collaboration

The Physical Environment Text produced so far

Composing medium

MOTIVATION / AFFECT

Beliefs & Attitudes Goals Predispositions Cost/benefit estimates WORKING MEMORY Phonological memory Semantic memory Visual/spatial sketchpad

(28)

14 Figure 2.4: Hayes' latest model of the writing process (2012, p.371) In its newest incarnation, Hayes' (2012) process level is split into writing processes and task environment. Writing processes are internal activities that a writer uses to compose his/her message, and include ‘proposer’ (non-verbal idea generation), ‘translator’ (transforms proposals into language), ‘transcriber’ (transforms the ‘translation’ into text), and ‘evaluator’ (assesses the output from the previous processes). Planning and revising (reviewing), then, are no longer separate processes in separate modules, but become 'a special application of the writing model' (p.376). In other words, the iterative nature of specialized activities (planning, summarizing, etc.) is now reconceptualised 'as declarative knowledge in LTM (as a stored plan or task scheme)' (p.375). In this view, activities that once reflected discrete cognitive processes are reconceptualised as parts of task schemas, which can be modified via instruction/experience. By placing the task environment and writing processes at the same level, this model also attempts to attend to how outside influences (physical, social, and cultural) may impact writing7_{. However,}

despite these improvements, the model still suffers from simplistic labelling, and much is left implicit in its description. For example, it is unclear as to how metalinguistic and metastrategic knowledge (presumably housed at the Resource Level) comes to be shaped by: (i) the eco-social elements of genre and intertextuality (presumably housed at the Control Level), and (ii) semiotic and material mediation (presumably housed at the Process Level). This is perhaps unsurprising given that there is little research into the role of metalinguistic knowledge in writing (Myhill, 2012,

7_{For example, research has frequently shown how the medium of transcription (typed vs. written) can}

(29)

15 p.253), or how eco-social demands (e.g., genre) link to writing processes. Furthermore, it is unclear how such a model would cope with the generation and maintenance of argumentative texts (texts with particulate patterns of meaning, cf. §3.3). As per other process models, there also appears to be little explicit discussion of how the organization of text may be constrained by the topic itself; for example, the chronological ordering of facts, or the logical relations between them, may impose constraints on the writer as to how they order ideas. Furthermore, in constructing argumentative text, the translator would be responsible for taking a complex multidimensional argument (represented in the mind of a writer as networked propositions and concepts) and transforming this gestalt representation into linearly segmented text. This process would involve the manipulation of complex lexicogrammar (cf. §3.3), anaphora (§3.4), and information structures (§3.5) to achieve rhetorical and pragmatic goals that align themselves with the audience and purpose of a text. Although such concerns are given theoretical status in Hayes' models (cf. the various labels used), they are rarely discussed—language, for example, is considered mainly in terms of syntax and lexis, with little mention of function (cf. Coirier et al., 1999, pp.16-18, for similar sentiments). Finally, with regards to the Control Level, although Hayes (2012) adds social and affective influences (e.g., Collaborators & Critics), motivation remains a blanket term for a whole host of variables. Consequently, the model still does not account for motivation in terms of its underlying constructs of self-efficacy, self-regulation, beliefs, etc. However, this lack of detailed attention to motivation may be an epiphenomenon of the research designs used in such paradigms: process research, for example, frequently uses experimental designs that call for texts to be produced for no other reason than providing data, while socio-constructivist research collects text(s) produced in authentic settings. It could be said, then, that much process research negates the need to look at motivation, while much socio-constructivist research takes motivation for granted8_.

Ultimately, while Hayes continues to update his model in line with current theorising, any model embedded in such a strong cognitive account of writing will inevitably be concerned with production rather than communication: writing as a skill rather than writing as a meaning-potential. For example, Leijten et al. (2014), in their case study of one professional writer spanning five writing episodes spread over four days, clearly illustrate how a complex writing task (business proposal) draws upon multiple sources (old and new) in the construction of a communicative and strategic goal. Using a combination of KSL, interviews, and online observations, they observed how one writer oriented to the reader (prospective client) by using external sources to retrieve content, structure and formulate text, design visuals and layout, and

8_{Academic literacies, for example, state that motivation (or lack thereof) to write is closely linked to a}

writer's self-belief in their ability to write, how they self-regulate cognition, affect, behaviour, and their environment, and to their underlying motive for writing (Hidi & Boscolo, 2007; Lea & Street, 1998).

(30)

16 provide additional warrant for claims. Using writing to construct new meaning from old is a form of discourse synthetic approach: ‘Readers become writers, creating new texts by selecting, organizing and connecting content from source texts’ (Spivey & King, 1989, p.7). Such an approach reflects a movement from knowledge-telling to knowledge-transforming (writing as a constructive process). However, the professional writer in Leijten, Van Waes, et al.’s (2014) study also assumed the reader’s point of view. This additional ‘developmental stage’ can be situated in relation to both Bereiter and Scardamalia (1987) and Kellogg (1996, 2008):

Figure 2.5: Macro-stages in the cognitive development of writing skill (Kellogg, 2008, p.4) Drawing parallels between high-level writing and the development of other high-level skills/abilities, such as playing chess or the violin, Kellogg (2008) suggests a developmental pathway for writing proficiency (as portrayed in figure 2.5), where writers move from knowledge-telling to knowledge-crafting after, on average, ten years of practice. In this view, a writer moves from intra- to inter-level concerns aided by cognitive maturity, disciplinary knowledge, and writing practice. However, although Kellogg's (2008) logic appears sound, it could be argued that some highly-skilled abilities are only mastered by those with a propensity for such talents. In conclusion, as Alamargot and Chanquoy (2001, p.21) note, most models subsume a number of cognitive components that are: (a) constrained by WM, (b) draw on 3 core activities (planning, formulating, and revising), and (c) controlled by a central management unit. Furthermore, writing expertise appears to stem from practice/maturity, which probably explains why the field as a whole sees writing as a 'skill' to be learnt (cf. the ubiquitous use of 'learning-to-write' in Manchón (2012) and Torrance et al. (2012); i.e. many process models assume a top-down approach to content integration, where block and arrow diagrams, whilst useful from a research design perspective, underrepresent and overlook the start-stop nature of composition. Fundamentally, any number of influential factors—changing task representations, human-transcription interface

Knowledge-telling Knowledge-transforming Knowledge-crafting ➊ Planning limited to idea

retrieval

➋ Limited interaction of planning and generation ➌ Minimal reviewing

➊ Interaction of planning, generating, and reviewing ➋ Reviewing of mainly author's representation Author Author Text Reader Author Text ➊ Interaction of planning, generating, and reviewing ➋ Reviewing of author and text representations Writ ing s ki ll Years of Practice 10 20

(31)

17 problems, gestation periods between drafts, (Daneman & Stainton, 1993), etc.—can dynamically alter the trajectory and quality of writing. Consequently, because writing is portrayed as an event, involving strategic and deliberate plan-translate-revise cycles in the pursuit of solving a 'problem', many models also assume text generation to be somewhat passive, and are thus too general and holistic in terms of the interpersonal and textual functions of language9_{. Furthermore, many}

models see text generation in terms of form (syntax and lexis), and this in turn negates the meaning-making potential of writing and side-lines rhetorical and pragmatic goals—aspects of language that are crucial to argumentative texts because it is the vehicle through which multidimensional structures (e.g., topics, arguments, and referents) are mapped onto linear sequences (Coirier et al., 1999).

2.2 Toward an integrated view of writing

Many of the models introduced above are based on writing as a problem solving activity. Consequently, they often carry an implicit assumption that writing is unidirectional—a process that begins in the mind and culminates in a text. However, more recent research views writing as both a problem solving endeavour (top-down) and a constructive process (bottom-up) (Galbraith, 2009a). This more modern view highlights the dialogic potential of writing, and thus accords more closely with view of writing assumed in this thesis—hence the inclusion of this section at this juncture. In this dialogic view of writing, writer and text, internal and external thought work together to advance text and the knowledge and ability of a writer. For example, in process research, there is the oft cited remark that writers typically adopt one of two polarised writing styles. These are deemed 'planners' and 'revisers': planners prefer to have a clear idea as to what they are writing before they start, and tend to write fewer drafts than revisers; revisers use writing to clarify their understanding, and tend to produce more drafts than planners. Torrance, Thomas, and Robinson (1994), however, intersect these two styles with an intermediary style deemed 'mixed strategy', which brings together the primarily top-down process of planning and the primarily bottom-up process of revision. Similar research premised on the existence of a cline of writing styles also reflects the duality of top-down/bottom-up processing (Kieft, Rijlaarsdam, & Van den Bergh, 2008). Perhaps the most well-known explication of this comes from Galbraith (1999, 2009a), and his research on self-monitoring. Galbraith posits a knowledge-constituting process, which aims to incorporate the emergence of knowledge that can occur during writing. This model draws heavily on the fact that text production is a major component of text generation (cf. Berninger & Swanson, 1994; Chenoweth & Hayes, 2001), and that text generation involves

9_{This problem is typically acknowledged in terms of 'pragmatics', but in keeping with the SFL basis of this}

Modelling choice in digital writing: functional revisions and 'texture'